with tf.name_scope("loss"):
    loss = tf.reduce_mean(tf.pow(tf.subtract(y, y_pred), 2))